Object Re-Identification Using Shared AI Features and Attributes

Solution Overview

Problem

Existing person re-identification systems face challenges in maintaining high accuracy and speed due to variations in illumination and attribute changes, and the separate operation of AI networks for representation and attribute extraction leads to inefficient resource usage.

Innovation Solution

An object re-identification apparatus that integrates AI networks for extraction and attribute extraction using a weight sharing method, and redefines the dataset based on attributes to improve performance speed and accuracy.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If separate AI networks are used for pedestrian re-identification representation extraction and attribute extraction, then the accuracy of re-identification technology is improved, but the performance speed becomes low and many GPU resources are used

Engineering Contradiction:

Improvere-identification accuracyVSAvoidperformance speed

Core Design Contradiction:

Measurement precisionVSProductivity

Solution Approach 1:

The patent merges the previously separate AI networks for pedestrian re-identification representation extraction and attribute extraction into a single integrated AI network. This network simultaneously performs both functions by sharing weights and parameters, reducing the number of independent models from two to one. The integration maintains the functional separation through distinct output branches while consolidating the feature extraction backbone, thereby improving performance speed and reducing GPU resource consumption while preserving re-identification accuracy.

Inventive Principle:

Principle #5Merging (Combining)

Solution Approach 2:

The integrated AI network is designed to perform multiple functions simultaneously: it extracts both pedestrian re-identification representations and attributes from the same input image through a unified architecture. The network uses shared weight parameters for feature extraction that benefit both tasks, making the system more efficient and resource-friendly while maintaining the capability to perform both re-identification and attribute extraction functions.

Inventive Principle:

Principle #6Universality (Multi-functionality)

2Measurement precision

If separate AI networks are used for pedestrian re-identification representation extraction and attribute extraction, then the accuracy of re-identification technology is improved, but many graphics processing unit (GPU) resources are used

Engineering Contradiction:

Improvere-identification accuracyVSAvoidGPU resource usage

Core Design Contradiction:

Measurement precisionVSUse of energy by moving object

Solution Approach 1:

The patent merges the previously separate AI networks for pedestrian re-identification representation extraction and attribute extraction into a single integrated AI network. This network simultaneously performs both functions by sharing weights and parameters, reducing the number of independent models from two to one. The integration maintains the functional separation through distinct output branches while consolidating the feature extraction backbone, thereby improving performance speed and reducing GPU resource consumption while preserving re-identification accuracy.

Inventive Principle:

Principle #5Merging (Combining)

Solution Approach 2:

The integrated AI network is designed to perform multiple functions simultaneously: it extracts both pedestrian re-identification representations and attributes from the same input image through a unified architecture. The network uses shared weight parameters for feature extraction that benefit both tasks, making the system more efficient and resource-friendly while maintaining the capability to perform both re-identification and attribute extraction functions.

Inventive Principle:

Principle #6Universality (Multi-functionality)

3Measurement precision

If existing re-identification technology is used, then it works in short-term re-identification situations where attribute does not change, but it has low accuracy in a situation where attribute changes

Engineering Contradiction:

Improvere-identification accuracy in stable conditionsVSAvoidaccuracy under attribute changes

Core Design Contradiction:

Measurement precisionVSAdaptability or versatility

Solution Approach 1:

The patent segments the feature extraction process into two distinct but integrated branches: one for extracting pedestrian re-identification representations and another for extracting attributes. This segmentation allows the system to handle attribute changes explicitly by comparing attribute features alongside representation features, improving accuracy in situations where attributes change (e.g., wearing different clothes or carrying items) while maintaining performance in stable conditions.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent introduces attribute parameters as additional features in the re-identification process. By incorporating explicit attribute extraction and comparison, the system adapts to parameter changes in the input data, such as changes in clothing, accessories, or carried items, thereby improving accuracy under varying conditions while maintaining robustness in stable scenarios.

Inventive Principle:

Principle #35Parameter changes

Data Source

PatentUS12561969B2Object re-identification apparatus and method thereof

Publication Date: 2026.02.24 HYUNDAI MOTOR CO LTD

AI summary

In an embodiment an apparatus includes a processor configured to generate a feature extraction module using a dataset in which an attribute for each object is defined, receive an image obtained by a camera, extract an attribute of an object of interest from the image using the learned feature extraction module, identify an object re-identification candidate group based on the extracted attribute of the object of interest and re-identify the object of interest based on the identified object re-identification candidate group.